Ethereum co-founder Vitalik Buterin has presented a demo version of an anonymous message board with moderation, deployed on the Aztec L2 network. The project's source code has already been published on GitHub, and this is not just a toy — it is a live prototype demonstrating how to combine complete anonymity of posts with effective content control.

The prototype is based on a concept Buterin first described back in 2022. He called the development a "toy version" created in vibe-coding mode, where 90-100% of the code is generated by AI agents. Despite the early stage, I see deep engineering meaning here: the mechanism allows a user to deposit ETH through a smart contract on the Ethereum mainnet and gain the ability to publish messages on Aztec, with funds withdrawable back to L1 at any time.

How does anonymity work?

The deposit address and its amount remain public, while withdrawing funds reveals the recipient's wallet on Ethereum. However, the messages themselves do not contain the sender's address in open data and are not linked to the original deposit. The anonymity set is formed from all users who have deposited funds but have not yet withdrawn them. This means privacy applies specifically to publications, not to all system operations — an important nuance to consider when assessing security.

The frequency of publications depends on the deposit size. With a base delay of 3600 seconds, the parameters are as follows:

  • 0.001 ETH — one message per hour;
  • 0.005 ETH — five messages per hour;
  • 0.01 ETH — ten messages per hour.

Unused limits can be accumulated and spent on multiple posts in a row. A repeat deposit does not bypass the restriction — a new initial delay is set for it. This is a well-thought-out mechanism to prevent abuse.

AI Moderation: Censorship without centralization?

When deploying the contract, a moderator address is assigned (the censor role). Its owner can mark messages as policy violations, after which they are hidden from the main feed. A user can view them through a separate warning, and the author receives an additional delay before the next message. The penalty size depends on a coefficient set during deployment.

The policy text is stored on-chain and limited to 1488 bytes. For automated checking, a local daemon with a large language model (LLM) is provided. The system works through llama.cpp and by default uses Qwen3.5-2B in a compressed format of about 1.4 GB. The model runs locally and does not access an external API — this ensures the moderation process does not depend on centralized services.

Moderator rights can be transferred to another address, including a multi-signature wallet or a DAO, but the protocol itself does not distribute authority automatically. This leaves room for flexible solutions but also requires careful management.

Technical limitations and risks

The repository is at an early stage. Some browser scenarios have not yet been tested. One known issue is the update_portal() function, which can be called by any user by specifying a fake address. This could lock the contract and render it unusable, although it does not allow funds to be stolen. Developers will need to strengthen zero-knowledge proofs and address other technical limitations.

My expert perspective

This prototype is not just an experiment. It demonstrates how zk-SNARKs technology and L2 solutions can create an environment where anonymity does not contradict order. AI moderation on a local model is a bold step that could become a standard for decentralized platforms. However, vulnerabilities such as contract lockup via update_portal() remind us that the path to production readiness is still long. Nevertheless, I believe it is precisely these "toy" versions that lay the foundation for future private yet regulated ecosystems.